FFmpeg
 All Data Structures Namespaces Files Functions Variables Typedefs Enumerations Enumerator Macros Groups Pages
dsputil.h
Go to the documentation of this file.
1 /*
2  * DSP utils
3  * Copyright (c) 2000, 2001, 2002 Fabrice Bellard
4  * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at>
5  *
6  * This file is part of FFmpeg.
7  *
8  * FFmpeg is free software; you can redistribute it and/or
9  * modify it under the terms of the GNU Lesser General Public
10  * License as published by the Free Software Foundation; either
11  * version 2.1 of the License, or (at your option) any later version.
12  *
13  * FFmpeg is distributed in the hope that it will be useful,
14  * but WITHOUT ANY WARRANTY; without even the implied warranty of
15  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16  * Lesser General Public License for more details.
17  *
18  * You should have received a copy of the GNU Lesser General Public
19  * License along with FFmpeg; if not, write to the Free Software
20  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
21  */
22 
23 /**
24  * @file
25  * DSP utils.
26  * note, many functions in here may use MMX which trashes the FPU state, it is
27  * absolutely necessary to call emms_c() between dsp & float/double code
28  */
29 
30 #ifndef AVCODEC_DSPUTIL_H
31 #define AVCODEC_DSPUTIL_H
32 
33 #include "libavutil/intreadwrite.h"
34 #include "avcodec.h"
35 #include "rnd_avg.h"
36 
37 /* encoding scans */
38 extern const uint8_t ff_alternate_horizontal_scan[64];
39 extern const uint8_t ff_alternate_vertical_scan[64];
40 extern const uint8_t ff_zigzag_direct[64];
41 extern const uint8_t ff_zigzag248_direct[64];
42 
43 /* pixel operations */
44 #define MAX_NEG_CROP 1024
45 
46 /* temporary */
47 extern uint32_t ff_squareTbl[512];
48 extern const uint8_t ff_cropTbl[256 + 2 * MAX_NEG_CROP];
49 
50 void ff_put_pixels8x8_c(uint8_t *dst, uint8_t *src, ptrdiff_t stride);
51 void ff_avg_pixels8x8_c(uint8_t *dst, uint8_t *src, ptrdiff_t stride);
52 void ff_put_pixels16x16_c(uint8_t *dst, uint8_t *src, ptrdiff_t stride);
53 void ff_avg_pixels16x16_c(uint8_t *dst, uint8_t *src, ptrdiff_t stride);
54 
55 /* RV40 functions */
56 void ff_put_rv40_qpel16_mc33_c(uint8_t *dst, uint8_t *src, ptrdiff_t stride);
57 void ff_avg_rv40_qpel16_mc33_c(uint8_t *dst, uint8_t *src, ptrdiff_t stride);
58 void ff_put_rv40_qpel8_mc33_c(uint8_t *dst, uint8_t *src, ptrdiff_t stride);
59 void ff_avg_rv40_qpel8_mc33_c(uint8_t *dst, uint8_t *src, ptrdiff_t stride);
60 
61 void ff_gmc_c(uint8_t *dst, uint8_t *src, int stride, int h, int ox, int oy,
62  int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
63 
64 /* minimum alignment rules ;)
65 If you notice errors in the align stuff, need more alignment for some ASM code
66 for some CPU or need to use a function with less aligned data then send a mail
67 to the ffmpeg-devel mailing list, ...
68 
69 !warning These alignments might not match reality, (missing attribute((align))
70 stuff somewhere possible).
71 I (Michael) did not check them, these are just the alignments which I think
72 could be reached easily ...
73 
74 !future video codecs might need functions with less strict alignment
75 */
76 
77 /* add and put pixel (decoding) */
78 // blocksizes for op_pixels_func are 8x4,8x8 16x8 16x16
79 //h for op_pixels_func is limited to {width/2, width} but never larger than 16 and never smaller than 4
80 typedef void (*tpel_mc_func)(uint8_t *block/*align width (8 or 16)*/, const uint8_t *pixels/*align 1*/, int line_size, int w, int h);
81 typedef void (*qpel_mc_func)(uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, ptrdiff_t stride);
82 
83 typedef void (*op_fill_func)(uint8_t *block/*align width (8 or 16)*/, uint8_t value, int line_size, int h);
84 
85 #define DEF_OLD_QPEL(name)\
86 void ff_put_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, ptrdiff_t stride);\
87 void ff_put_no_rnd_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, ptrdiff_t stride);\
88 void ff_avg_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, ptrdiff_t stride);
89 
90 DEF_OLD_QPEL(qpel16_mc11_old_c)
91 DEF_OLD_QPEL(qpel16_mc31_old_c)
92 DEF_OLD_QPEL(qpel16_mc12_old_c)
93 DEF_OLD_QPEL(qpel16_mc32_old_c)
94 DEF_OLD_QPEL(qpel16_mc13_old_c)
95 DEF_OLD_QPEL(qpel16_mc33_old_c)
96 DEF_OLD_QPEL(qpel8_mc11_old_c)
97 DEF_OLD_QPEL(qpel8_mc31_old_c)
98 DEF_OLD_QPEL(qpel8_mc12_old_c)
99 DEF_OLD_QPEL(qpel8_mc32_old_c)
100 DEF_OLD_QPEL(qpel8_mc13_old_c)
101 DEF_OLD_QPEL(qpel8_mc33_old_c)
102 
103 /* motion estimation */
104 // h is limited to {width/2, width, 2*width} but never larger than 16 and never smaller than 2
105 // although currently h<4 is not used as functions with width <8 are neither used nor implemented
106 typedef int (*me_cmp_func)(void /*MpegEncContext*/ *s, uint8_t *blk1/*align width (8 or 16)*/, uint8_t *blk2/*align 1*/, int line_size, int h)/* __attribute__ ((const))*/;
107 
108 /**
109  * Scantable.
110  */
111 typedef struct ScanTable{
113  uint8_t permutated[64];
114  uint8_t raster_end[64];
115 } ScanTable;
116 
117 void ff_init_scantable(uint8_t *, ScanTable *st, const uint8_t *src_scantable);
118 void ff_init_scantable_permutation(uint8_t *idct_permutation,
119  int idct_permutation_type);
120 
121 /**
122  * DSPContext.
123  */
124 typedef struct DSPContext {
125  /* pixel ops : interface with DCT */
126  void (*get_pixels)(int16_t *block/*align 16*/, const uint8_t *pixels/*align 8*/, int line_size);
127  void (*diff_pixels)(int16_t *block/*align 16*/, const uint8_t *s1/*align 8*/, const uint8_t *s2/*align 8*/, int stride);
128  void (*put_pixels_clamped)(const int16_t *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
129  void (*put_signed_pixels_clamped)(const int16_t *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
130  void (*add_pixels_clamped)(const int16_t *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
131  void (*add_pixels8)(uint8_t *pixels, int16_t *block, int line_size);
132  int (*sum_abs_dctelem)(int16_t *block/*align 16*/);
133  /**
134  * translational global motion compensation.
135  */
136  void (*gmc1)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x16, int y16, int rounder);
137  /**
138  * global motion compensation.
139  */
140  void (*gmc )(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int ox, int oy,
141  int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
142  void (*clear_block)(int16_t *block/*align 16*/);
143  void (*clear_blocks)(int16_t *blocks/*align 16*/);
144  int (*pix_sum)(uint8_t * pix, int line_size);
145  int (*pix_norm1)(uint8_t * pix, int line_size);
146 // 16x16 8x8 4x4 2x2 16x8 8x4 4x2 8x16 4x8 2x4
147 
148  me_cmp_func sad[6]; /* identical to pix_absAxA except additional void * */
162 
167  me_cmp_func ildct_cmp[6]; //only width 16 used
168  me_cmp_func frame_skip_cmp[6]; //only width 8 used
169 
170  int (*ssd_int8_vs_int16)(const int8_t *pix1, const int16_t *pix2,
171  int size);
172 
173  /**
174  * Thirdpel motion compensation with rounding (a+b+1)>>1.
175  * this is an array[12] of motion compensation functions for the 9 thirdpe
176  * positions<br>
177  * *pixels_tab[ xthirdpel + 4*ythirdpel ]
178  * @param block destination where the result is stored
179  * @param pixels source
180  * @param line_size number of bytes in a horizontal line of block
181  * @param h height
182  */
183  tpel_mc_func put_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
184  tpel_mc_func avg_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
185 
190 
192 
193  /* huffyuv specific */
194  void (*add_bytes)(uint8_t *dst/*align 16*/, uint8_t *src/*align 16*/, int w);
195  void (*diff_bytes)(uint8_t *dst/*align 16*/, const uint8_t *src1/*align 16*/, const uint8_t *src2/*align 1*/,int w);
196  /**
197  * subtract huffyuv's variant of median prediction
198  * note, this might read from src1[-1], src2[-1]
199  */
200  void (*sub_hfyu_median_prediction)(uint8_t *dst, const uint8_t *src1, const uint8_t *src2, int w, int *left, int *left_top);
201  void (*add_hfyu_median_prediction)(uint8_t *dst, const uint8_t *top, const uint8_t *diff, int w, int *left, int *left_top);
202  int (*add_hfyu_left_prediction)(uint8_t *dst, const uint8_t *src, int w, int left);
203  void (*add_hfyu_left_prediction_bgr32)(uint8_t *dst, const uint8_t *src, int w, int *red, int *green, int *blue, int *alpha);
204  /* this might write to dst[w] */
205  void (*bswap_buf)(uint32_t *dst, const uint32_t *src, int w);
206  void (*bswap16_buf)(uint16_t *dst, const uint16_t *src, int len);
207 
208  void (*h263_v_loop_filter)(uint8_t *src, int stride, int qscale);
209  void (*h263_h_loop_filter)(uint8_t *src, int stride, int qscale);
210 
211  /* assume len is a multiple of 8, and arrays are 16-byte aligned */
212  void (*vector_clipf)(float *dst /* align 16 */, const float *src /* align 16 */, float min, float max, int len /* align 16 */);
213 
214  /* (I)DCT */
215  void (*fdct)(int16_t *block/* align 16*/);
216  void (*fdct248)(int16_t *block/* align 16*/);
217 
218  /* IDCT really*/
219  void (*idct)(int16_t *block/* align 16*/);
220 
221  /**
222  * block -> idct -> clip to unsigned 8 bit -> dest.
223  * (-1392, 0, 0, ...) -> idct -> (-174, -174, ...) -> put -> (0, 0, ...)
224  * @param line_size size in bytes of a horizontal line of dest
225  */
226  void (*idct_put)(uint8_t *dest/*align 8*/, int line_size, int16_t *block/*align 16*/);
227 
228  /**
229  * block -> idct -> add dest -> clip to unsigned 8 bit -> dest.
230  * @param line_size size in bytes of a horizontal line of dest
231  */
232  void (*idct_add)(uint8_t *dest/*align 8*/, int line_size, int16_t *block/*align 16*/);
233 
234  /**
235  * idct input permutation.
236  * several optimized IDCTs need a permutated input (relative to the normal order of the reference
237  * IDCT)
238  * this permutation must be performed before the idct_put/add, note, normally this can be merged
239  * with the zigzag/alternate scan<br>
240  * an example to avoid confusion:
241  * - (->decode coeffs -> zigzag reorder -> dequant -> reference idct ->...)
242  * - (x -> reference dct -> reference idct -> x)
243  * - (x -> reference dct -> simple_mmx_perm = idct_permutation -> simple_idct_mmx -> x)
244  * - (->decode coeffs -> zigzag reorder -> simple_mmx_perm -> dequant -> simple_idct_mmx ->...)
245  */
248 #define FF_NO_IDCT_PERM 1
249 #define FF_LIBMPEG2_IDCT_PERM 2
250 #define FF_SIMPLE_IDCT_PERM 3
251 #define FF_TRANSPOSE_IDCT_PERM 4
252 #define FF_PARTTRANS_IDCT_PERM 5
253 #define FF_SSE2_IDCT_PERM 6
254 
255  int (*try_8x8basis)(int16_t rem[64], int16_t weight[64], int16_t basis[64], int scale);
256  void (*add_8x8basis)(int16_t rem[64], int16_t basis[64], int scale);
257 #define BASIS_SHIFT 16
258 #define RECON_SHIFT 6
259 
260  void (*draw_edges)(uint8_t *buf, int wrap, int width, int height, int w, int h, int sides);
261 #define EDGE_WIDTH 16
262 #define EDGE_TOP 1
263 #define EDGE_BOTTOM 2
264 
265  void (*shrink[4])(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
266 
267  /**
268  * Calculate scalar product of two vectors.
269  * @param len length of vectors, should be multiple of 16
270  */
271  int32_t (*scalarproduct_int16)(const int16_t *v1, const int16_t *v2/*align 16*/, int len);
272  /* ape functions */
273  /**
274  * Calculate scalar product of v1 and v2,
275  * and v1[i] += v3[i] * mul
276  * @param len length of vectors, should be multiple of 16
277  */
278  int32_t (*scalarproduct_and_madd_int16)(int16_t *v1/*align 16*/, const int16_t *v2, const int16_t *v3, int len, int mul);
279 
280  /**
281  * Apply symmetric window in 16-bit fixed-point.
282  * @param output destination array
283  * constraints: 16-byte aligned
284  * @param input source array
285  * constraints: 16-byte aligned
286  * @param window window array
287  * constraints: 16-byte aligned, at least len/2 elements
288  * @param len full window length
289  * constraints: multiple of ? greater than zero
290  */
291  void (*apply_window_int16)(int16_t *output, const int16_t *input,
292  const int16_t *window, unsigned int len);
293 
294  /**
295  * Clip each element in an array of int32_t to a given minimum and maximum value.
296  * @param dst destination array
297  * constraints: 16-byte aligned
298  * @param src source array
299  * constraints: 16-byte aligned
300  * @param min minimum value
301  * constraints: must be in the range [-(1 << 24), 1 << 24]
302  * @param max maximum value
303  * constraints: must be in the range [-(1 << 24), 1 << 24]
304  * @param len number of elements in the array
305  * constraints: multiple of 32 greater than zero
306  */
308  int32_t max, unsigned int len);
309 
311 } DSPContext;
312 
313 void ff_dsputil_static_init(void);
314 void ff_dsputil_init(DSPContext* p, AVCodecContext *avctx);
317 
318 int ff_check_alignment(void);
319 
320 void ff_set_cmp(DSPContext* c, me_cmp_func *cmp, int type);
321 
329 
331 
332 #endif /* AVCODEC_DSPUTIL_H */